Ali,
I agree to
your statement “single channel SFP+ is hard and
doing the same with 10 channel will only be harder”. Do you
think the 4-channel solution will be as hard as the
single channel or as hard as the 10-channel? If it is the later, or
close to
the later, shouldn’t we simply jump onto the 10-channel and forget the
4-channel?
Wenbin
Chris
It would be difficult to explain everything in an email, so I am going
to be
short and probably do a presentation in July.
Please see my comments below in blue.
Chris Cole wrote:
Ali,
It is not clear what problems are solved by your proposal to define the 40G and 100G MMF optical specifications around linear interfaces and host EDC.
Fist, SFP+ 10GE-SR optics are a solved problem today, and the SFP+ 10GE-SR specification, while not optimum, is complete and manufacturable. The original 300m meter 802.3 10GE-SR specification has a number of issues which affect yield and therefore delayed the availability of low cost 10GE. However, as has been seen from numerous emails on this reflector, multiple manufacturers have resolved these issues.
As
the editor of SFP+ I would say we are
working very hard to solve the 10G-SR and LRM SFP+. I can also tell
you
the limiting interface is the hardest one
due to jitter budget. You can also download the latest SFP+
specifications from the SFF website.
ftp://ftp.seagate.com/sff/SFF-8431.PDF
On the other hand, the SFP+ 10GE-LRM linear specification still has a number of difficult issues to resolve. So you are proposing to consider as a starting point a spec which is still under discussion and therefore not done, instead of a specification which is complete and verified to work.
LRM uses a
linear interface and some of the difficulty is
due to TWDP limits and has nothing to do with linear interface. You
need
to separate linear
interface benefit from LRM difficulties.
The reason for going to a lower distance for 40G and 100G is to provide additional margin/yield. Further, 300m multi-ribbon applications are highly unlikely, so it makes little sense to have the 40G or 100G specifications driven by an insignificant fraction of the applications. We may re-visit whether 100m is the right distance (150m has been mentioned as an alternative,) but 300m would be of little value as was commented on by multiple-end users during the HSSG discussion of the MMF objective.
Second, I am at loss as to how an EDC solves the additional penalty issue of cross-talk in a multi-lane application. Cross-talk problems are solved through echo-cancellers, not EDCs. So if we wanted to reduce cross-talk effects through signal processing techniques, a solution resembling 1000BASE-T or 10GBASE-T would be required. 10GBASE-T power numbers in the many watts have been reported on this reflector. For 40G, we would expect linear scaling in power, and quadratic scaling for full-cross ten-lane echo-cancellation. We will be well served to view this as a solution of last resort, not as a starting point.
I
did not say solve the XTALK issue with
EDC or possibly use echo-cancelers. I was saying due to XTALK,
additional
channel impairments when routing
10 channels, and SerDes with more jitter you will need to either put a
retimer
in the module or use a SerDes with build in EDC as better compromise.
Third, I do not see the motivation to have an EDC for an objective which explicitly states OM3 as the fiber. OM3 does not have dispersion problems over a distance of 100m or 150m. LRM EDC was developed for legacy OM1 fiber, already deployed within buildings, for example between floors. I have heard no application identified in any HSSG presentation for 40G or 100G which would use ribbon-fiber that had dispersion problems like OM1. So we would burden 40G or 100G hosts with an EDC per channel, so that we can use optics that do not meet SR specs on the speculatively assumption that they are lower cost.
I
was not saying to use the power of EDC
to solve fibre dispersion, but rather to solve higher transmit jitter
contribution, laser BW limitation, some fiber dispersion,
and the electrical receive chain.
Fourth, cross-talk for connectors and PCB traces has been simulated and quantified (see for example page 12 of cole_01_1106.pdf.) There is no indication that the cross-talk magnitude is anywhere near requiring the drastic measures of an EDC/Echo Canceller. A careful re-allocation of the SFI (SFP+ interface) jitter budget between the host and optics will permit tolerance of these levels of cross-talk.
I
looked at this presentations, on the
SFP+ Module Compliance Boards Broadcom is building for the industry
with lots
of care the
NEXT measurements are attached. We do measure more crosstalk than your
theoretical limits. Chris it does not matter how careful you
are here is the facts:
- 10X SerDes transmitter will have more jitter
- More NEXT
- 10 channel optics performance may not be as good as single
channel
- 10X SerDes receiver will have less jitter tolerance
with limiting interface you only have 1 UI to work with but using
linear
interface you can stretch your 1 UI!
Missing is the measurement data for cross-talk in multi-lane 10G I/O CMOS ASICs. Until we have solid data for this, we will not be able to complete the specification of MMF PMDs that do not require CDRs.
I am
telling you single channel SFP+ is
hard and doing the same with 10 channel will only be harder.
We also need to set a realistic expectation and not to over promise and
under
deliver!
Thanks,
Ali
I would encourage all IC vendors participating in the HSSG, who have developed silicon that implements 10G I/O, to bring in multi-lane 10G I/O cross-talk data so that we can base the 40G and 100G specifications on measurement results.
Chris
----------------------------------------------------------------------------
"Ali Ghiasi" <aghiasi@broadcom.com>
06/27/2007 12:33 PM
To: "Jack Jewell" <Jack.Jewell@PICOLIGHT.COM>, "Paul Kolesar" <pkolesar@systimax.com>
cc: STDS-802-3-HSSG@listserv.ieee.org
Subject: Re: [HSSG] The List
Jack and Paul
The question is not whether SFP+ can achieve 300 m SR reach similar to XFP, but how do we get to 10G SFP+
at 2.5x the cost of 1G classic SFP for DCE (Data Center Ethernet) with max reach of 100 m.
If we can get to 10G SFP+ at 2.5x the cost of 1G at 300 m then the 10G PAR objective is complete, but how long
do we wait the need is know. But I do know the combination of lower cost optics with EDC can deliver
the 2.5x cost objective for DCE applications near term. To get to these cost the transmitter very likely
will not be fully SR compliant and in that case it does not matter if the reach is 100 or 300 m.
The current assumption in the HSSG is that you can achieve SFP+ limiting performance with 4 or 10 channels without the
use of CDR in the module, with more crosstalk, less optimum layout, SerDes having more jitter and less tolerance compare
to small port count PHYs, optics ??? You will get small benefit from reducing fiber reach to 100m but not enough to close the
link budget. As Dan mentioned EDC is becoming a standard feature on PHYs and we definitely need to leverage it for 40G/100G.
Use of linear interface is an approach that can close the link budget without the use of CDR in the module, relax the optics
specifications, and the same interface can support passive copper Twin-ax up to 10m.
Ali